introduction of intel extension for pytorch

[jingxu10]

PR to add introduction of Intel Extension for PyTorch to PyTorch tutorials.

[netlify]

✔️ Deploy Preview for pytorch-tutorials-preview ready!

🔨 Explore the source changes: 7b0dc38

🔍 Inspect the deploy log: https://app.netlify.com/sites/pytorch-tutorials-preview/deploys/615e9072f7792800074087b3

😎 Browse the preview: https://deploy-preview-1702--pytorch-tutorials-preview.netlify.app/recipes/recipes/intel_extension_for_pytorch

[Jianhui-Li]

@vitaly-fedyunin @gottbrath This is the IPEX tutorial.

[malfet]

Two minor typos

[yqhu]

Tested the inference examples (Imperative and Script modes; Float32 and BFloat16) with IPEX built from master and model = ipex.optimize(model, ...) triggered the assertion on https://github.com/intel/intel-extension-for-pytorch/blob/master/torch_ipex_py/utils.py#L141

Traceback (most recent call last):
  File "inf_fp16.py", line 17, in 
    model = ipex.optimize(model, dtype=torch.bfloat16)
  File "/home/ubuntu/anaconda3/envs/ipex/lib/python3.8/site-packages/intel_extension_for_pytorch/utils.py", line 141, in optimize
    assert optimizer is not None, "The optimizer should be given for training mode"
AssertionError: The optimizer should be given for training mode

Was curious why model.training would be true for inference. Or did I build from the wrong branch of IPEX?

The build process generated libintel_pex.so so I had to change libintel-ext-pt-cpu.so to libintel_pex.so in CMakeLists.txt for the C++ example to compile. Again, it could be me using the wrong branch?

[jingxu10]

Hi @yqhu , model.eval() is required if it is an inference workload. Since we just would like to show code changes to the original pytorch code, I didn't show full-functioning code snippets. Please let me know if you prefer to having full code in the tutorial.
We were under discussion about the c++ dynamic library name recently. The library name in the current master branch will be changed to libintel-ext-pt-cpu.so later.

[yqhu]

model.eval() won't fix it though I understand it's still under development. Thanks.

[jingxu10]

Hi @yqhu , I fixed the AssertionError: The optimizer should be given for training mode error in the latest commit. Pls have a try.

[yqhu]

Thanks @jingxu10.

[gottbrath]

Just wanted to chime in here and say that I think there is value in having this tutorial be part of the PyTorch tutorial set.